Skip to content
datanalyze Statistics & data science
Language FR EN
Get in touch
Services Training About Get in touch
Language FR EN

Service

Study design and sample size

This is the only point at which everything is still fixable. I help you size your study, choose the right variables and write the analysis plan — before collection locks in your options.

Discuss your project
01

Who it is for

This service comes in up front, when the data does not exist yet. It is also the least requested, even though it is the one that prevents the most damage: an under-sized study cannot be rescued afterwards.

  • Clinical researchers submitting a protocol
  • PhD students at the start of their thesis
  • Teams preparing an ethics committee application
  • Applicants who must justify a funding request
  • Companies about to run a survey or an A/B test
  • Labs planning a measurement campaign
02

The problem it solves

Most analyses that fail do not fail at the analysis stage. They fail at collection: too few subjects to detect the effect sought, a confounding variable nobody thought to measure, two groups formed in a way that makes them no longer comparable.

These mistakes have an unpleasant property: they are irreversible. No statistical method, however sophisticated, recovers information that was never collected. You can only note the damage and publish an inconclusive result.

A few hours of design up front cost a fraction of a collection redone — when redoing it is even possible.

  • An ethics committee wants the sample size justified
  • You do not know how many subjects to include
  • You are unsure which variables to measure
  • You have to write the "statistical analysis" section of a protocol
  • A funder asks for the expected statistical power
  • You want to avoid repeating the last study's mistake
03

What you get

Deliverables are written to drop straight into your protocol, your ethics application or your funding request.

  • The sample size calculation, with its assumptions spelled out
  • The power analysis and alternative scenarios
  • A written, dated and versioned statistical analysis plan
  • The list of variables to measure, and why each one
  • Recommendations on group allocation and randomisation
  • A "statistical methods" section ready for your protocol
04

How it works

  1. Research question

    We start from what you want to demonstrate and turn it into a testable hypothesis. That step determines everything else.

  2. Feasibility

    How many subjects can you realistically recruit, in what time, on what budget? The constraints are part of the calculation.

  3. Sizing

    Sample size computed under several effect scenarios, so you choose with full knowledge of the trade-off.

  4. Analysis plan

    The plan and the methods section written up, ready to attach to your application.

05

Concrete examples

Methodological illustrations of the questions handled at this stage.

How many patients to include?

Detecting a clinically meaningful difference with 80% power: the calculation, its assumptions, and what happens if recruitment falls short.

Which variables to measure?

Identifying the confounders to collect up front. A variable missed at this stage can never be added to the model later.

How to form the groups?

Simple, stratified or block randomisation — and what each choice implies for the analysis that follows.

A usable questionnaire

Wording the questions and response scales so that the resulting data can actually be analysed.

Sizing an A/B test

How many visitors and how long before you can conclude, rather than stopping the test as soon as the curve looks good.

Interim analyses

Planning to look at results mid-study without invalidating the final test: the rules are set beforehand, not after.

Frequently asked questions

Is this really necessary for a small study?
Small studies are exactly where it matters most. With a large sample, an approximate method often remains usable; with thirty subjects, poor sizing condemns the study to conclude nothing at all. The result is then not “negative”, it is uninterpretable — which is far worse.
I have no idea what effect to expect. How can a sample size be calculated?
That is the normal situation and it is not a blocker. We work in scenarios: what is the smallest effect that would genuinely matter to you? We size on that value, and I show you how the number of subjects moves with the assumptions. You then choose knowing the trade-off between recruitment cost and the risk of detecting nothing.
Collection has already started. Is it too late?
Not necessarily. Depending on how far along you are, it is often still possible to adjust recruitment, add a variable or fix the analysis plan. Get in touch: I will tell you frankly what can still be salvaged and what cannot. If analysis is your real need, data analysis or support for researchers is the better fit.
Will the ethics committee accept the calculation?
A sample size calculation is accepted when its assumptions are explicit and justified: expected effect and its source, standard deviation, significance level, target power, planned test. That is exactly what the deliverable contains, written to go straight into your application.
How long does it take?
It is the shortest service in the catalogue: most requests are settled in one or two working sessions, once the research question is clear. Set against what it prevents, it is the best time-to-benefit ratio in the catalogue.

Other services

Data analysis

You have the data, I produce the results

Learn more

Support for researchers

You run the analysis, I secure the method

Learn more

Consulting and decision support

From dormant data to evidence-based decisions

Learn more

R development and automation

An analysis that runs itself, identically

Learn more

Training

Making your teams self-sufficient in R and data analysis

Learn more

A dataset to analyse, a method to validate?

Describe your need in a few lines, or book a first no-commitment call. I will tell you straight if I am the right person for it.

Describe my need Book a call